Even if we assume that there is a strong correlation between IQ and intelligence, there are still some problems, especially with the measurement and interpretation of high IQ results.
Most prominently, pretty much all widely used IQ tests cap their results, typically at three standard deviations (3σ, 3 sigma). Three standard deviations cover almost 99.9% of the sample (if you discard extremely low scores), i.e. 999 out of 1000 people. [f2] This looks pretty neat until you scale it up to the size of the world's population. Then you see that you miss 9.9 million people world-wide who have an IQ beyond +3σ. [f3]
Some modern tests extend their range to +4σ, which covers 99.997% of the population. This is a big improvement (leaving "only" about 232,000 people in the dust.) [f4]
It has been argued by some psychologists that there is not really a difference between a +2σ and +4σ IQ, because they are both "above average". This point of view is probably based on the fact that IQ tests are mostly used to identify mentally challenged people, while a high IQ is not associated with any social or psychological difficulties. Unfortunately, this point of view is not congruent with the experiences made by many highly intelligent people.
Another problem with high scores is that they often get measured inaccurately, because little research is performed in the area of high IQ and little attention is paid to raw data. (Some tests are even evaluated automatically, so nobody ever has a look at the underlying data.)
When a test score is "high", though, it often makes sense to look at the raw data. When the data show significant "peaks" or "valleys", for instance, the result might be inaccurate. Maybe the test subject even fell asleep from boredom while taking the test (yes, such things do happen). 
Finally, measuring very high IQs (beyond 4σ) is tricky, because people with such high IQs are rare (about 1 out of 31,457), [f5] so it is hard to find suitable test subjects for the verification of the tests. Also, there is always the risk that the subject is "too smart" for the test.
This is not a carte blanche for people "failing" tests, though! If you score low in an IQ test, chances of you being too smart are more than 1:30,000 against your favor. [f5] Most highly intelligent people still score reasonably high (2..3σ) in most tests and/or their raw scores exhibit significant anomalies.